Evaluating strategies for achieving and maintaining positive lifestyle behavior changes is a key focus of public health research studies. However, assessing self-reported behavioral measurements through statistical modeling is challenging because of their measurement errors due to gaps and biases in participant reporting, as well as the multidimensionality of interrelated measurements. Disease Biomarkers may provide a proximal measure of dietary intake, but the lack of specific dietary biomarkers has been recognized as an area requiring future research. Moreover, the measurement of biomarkers is often subject to left-censoring due to detection limits. We propose a statistical approach that constructs a quantile-specific mixture of multiple dietary intake components. Under the quantile regression framework, the proposed method characterizes dynamic patterns of multiple food items and evaluates their relative effects on varying levels of disease risk determined by biomarker levels, while accounting for both complex correlation structures of multiple behaviors and censoring issues in longitudinal biomarker data. We also apply the method to data from a behavioral intervention trial.